Dear SPMers,
I’m currently involved in the analysis of an event-related fMRI study with a complex design & will need some advice. The design has a total of 4 factors (2 between-, 2 within-subject). Two factors are within-subject – given these repeated measures & previous posts I think choosing the flexible factorial design (instead of the full factorial) might reveal an analysis with a higher sensitivity. Correct?
Moreover, using this approach the main- & interaction-effects will be modeled using specific contrasts from the 1st level. The factors are:
(1) Group (between; patients vs. controls)
(2) Participants gender (between; male vs. female)
(3) Emotional face category (within; happy, angry, fear, sad, disgust, neutral)
(4) Emotional face gender (within; male vs. female)
To address the specific effects on the 2nd level I would sum up across the within subject factors on the 1sdt level & produce:
1. Contrasts for each emotional category X gender (e.g. happy_male, happy_female….)
2. Contrasts for each emotional category across gender (e.g. happy, angry,…)
3. Contrasts for male & female stimuli across emotion (all female faces, all male faces)
4. Contrast that sums up across all conditions (to later assess the simple main effect of group)
Correct?
The effects on the second level will then be modeled using the flexible factorial design & t-tests on the specific contrasts. However, I would not be able to model all factors in one model & some argue that a model with all factors is needed to show the overall significance of the model (using an omnibus F-test), and that this is a necessary requirement for using reduced models on specific interaction effects. Is this a statistical requirement?
Thanks very much in advance & best regards,
Ben
|